A Multimodal deep learning method for predicting drug-induced liver injury using structural and sequential molecular...

Area of Science:

• Computational toxicology
• Pharmacology
• Machine learning in drug discovery

Background:

• Drug-induced liver injury (DILI) is a major challenge in drug development, leading to failures and withdrawals.
• Current computational models often overlook sequential information in SMILES, focusing primarily on structural features.

Purpose of the Study:

• To develop a multimodal deep learning framework for enhanced DILI prediction.
• To integrate both substructural and sequential molecular information for improved accuracy.

Main Methods:

• Utilized diverse molecular fingerprints (KlekotaRoth, MACCS, ECFP2, etc.) for structural features.
• Employed advanced language models (ChemBERTa, RoBERTa, SMILES2Vec) for sequential SMILES data.
• Developed a multimodal deep learning framework fusing these complementary data modalities.

Main Results:

• The multimodal approach demonstrated superior performance over single-modality methods.
• An ablation study confirmed the synergistic contribution of both structural and sequential features.
• The model effectively captures complex biological mechanisms underlying liver toxicity.

Conclusions:

• Multimodal integration is crucial for developing accurate and generalizable DILI predictive models.
• This framework offers a promising direction for improving drug safety assessment.
• The findings highlight the value of combining diverse data types in computational toxicology.